Abstract
Understanding the interactions among agricultural processes, soil, and plants is necessary for optimizing crop yield and productivity. This study focuses on developing effective monitoring and analysis methodologies that estimate key soil and plant properties using unmanned aerial vehicles (UAVs) and surface geophysical techniques. Applied to a soybean farm in Arkansas, these approaches characterize soil-plant coupled spatial and temporal heterogeneity. UAV-based multitemporal RGB imagery was used to monitor plant height and photosynthetic activity, while an algorithm estimated plant spatial abundance and vigor. High-resolution surface geophysical measurements of apparent soil electrical conductivity ($EC_a$) estimated soil texture heterogeneity. By integrating these datasets, the study identified spatiotemporal co-variance between soil properties and plant development. Early season monitoring identified low-productivity areas controlled by soil clay content, and temporal analysis showed the impact of soil moisture and irrigation (controlled by topography) on plant dynamics.
Introduction
In precision agriculture, the goal is to manage crops at the "micro-scale"—treating specific parts of a field differently based on their specific needs. However, most farmers rely on satellite imagery, which is often too coarse (2–5 meters per pixel) and frequently obscured by cloud cover. Furthermore, focusing solely on above-ground plant health ignores the "hidden" driver of productivity: the soil. This research addresses the challenge of characterizing soil-plant co-variability, investigating how heterogeneous soil properties—specifically clay content and moisture—influence soybean development from emergence to harvest.
Methods
The researchers developed a multi-modal monitoring framework consisting of three main pillars:
- UAV-Based Plant Analytics: Using a DJI Matrice 600 equipped with an RGB camera, "Structure from Motion" (SfM) photogrammetry was used to create 3D Digital Elevation Models (DEMs) for measuring plant height and 2D mosaics for calculating plant spatial abundance and vigor (via the Green Chromatic Coordinate, GCC).
- Ground-Based Geophysics: Electromagnetic Induction (EMI) measured apparent soil electrical conductivity ($EC_a$). As $EC_a$ is sensitive to soil texture (clay vs. sand) and moisture, it serves as a non-invasive tool to map subsurface conditions.
- Machine Learning Classification: To distinguish plants from soil in high-resolution (10 cm) imagery, a Support Vector Machine (SVM) algorithm was combined with mathematical morphology to capture the geometric patterns of crop rows.
Datasets and Study Site
The study was conducted on a 20-hectare soybean farm in the Arkansas Delta during the 2018 growing season. The field was instrumented with various sensors and surveyed multiple times to capture temporal dynamics.
The dataset included five UAV flights from May to September, four EMI surveys mapping the field's subsurface, and extensive ground truth data including 31 manual plots, 12 physical soil samples, and in-situ moisture/temperature sensors. High-resolution harvest data was collected directly from a combine harvester's onboard sensors to validate yield predictions.
Results and Discussion
The integration of multi-scale soil and plant datasets revealed several critical insights into crop productivity drivers.
The "Clay Trap" and Topographic Influence
A strong negative correlation was found between soil $EC_a$ and plant development. Areas with high clay content (high $EC_a$) exhibited poor drainage. During the wet early season, these areas became saturated, limiting oxygen to the roots and stunting growth.
Topography played a significant role in irrigation effectiveness. The field was divided into East, Center, and West sections. While the East was higher and closer to the water source, the West was lower. Interestingly, the West section became drier in mid-summer because furrow irrigation, controlled by the slope, was less effective there. This caused a shift in plant response; in July, the higher water-retention capability of clay-rich soil in the West became a temporary benefit, though overall yield remained lower due to initial drainage issues.
Early Detection and Yield Prediction
The UAV-based plant abundance algorithm showed strong agreement with ground measurements ($R^2 = 0.82$). By using 10-cm resolution imagery, "sprout density" (spatial abundance) could be estimated very early in the season. This metric proved to be a remarkably accurate predictor of final yield, suggesting that farmers could identify anomalous areas weeks earlier than with traditional methods. While vigor (greenness) and abundance (coverage) are related, they provide distinct information: abundance captures the crop "stand" density, while vigor reflects physiological health.
Conclusion
This study demonstrates that coupling UAV data products with geophysical imagery provides critical information for farm management. While the high-resolution data stream is labor-intensive, integrating UAV data with satellite imagery offers a potential path for scaling these insights to larger industrial farms. Future work should explore different crop types and the timing of UAV flights relative to specific growth stages to avoid misleading indices during late-season senescence.
Summary updated